BloomBench - Arabic - Understand: leaderboard

Metric: Accuracy (%) on the 1592 Arabic BloomBench items at the Bloom level Understand (compositional and cognitive understanding: relations, emotions, semantics), four-option image-question multiple choice, zero-shot, temperature 0, the chosen letter parsed from the generated answer (Regex-based Answer Extraction; unparseable answers count as wrong); chance 25; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.

Top models

#ModelScore
1Gemma 3 27B93
2Gemma 3 12B92
3Gemma 4 31B91
4Gemma 4 26B A4B89
5Qwen 2.5 VL 7B87
6GPT-4o Mini86
7Qwen 2 VL 7B86
8Gemma 3 4B84

Interactive version: theaggregate.ai/benchmark?slug=bloombench-arabic-understand · How It Works · Data refreshed daily, snapshot 2026-09-29.